Online Appendix for Does mobile money affect saving behavior? Evidence from a developing country Journal of African Economies

Transcription

1 Online Appendix for Does mobile money affect saving behavior? Evidence from a developing country Journal of African Economies Serge Ky, Clovis Rugemintwari and Alain Sauviat In this document we report the following appendices: O1. External validity O2. Impact of mobile money on individuals saving behavior using an alternative source of data Table O1. Access to formal financial services by region in Burkina Faso. Table O2. Saving choices and mobile money: using Global Financial Inclusion Database. Table O3. Transaction fees of mobile money services. Fees associated with mobile money cash in / cash out functions and transfers services as of Table O4. Distance to the nearest mobile money agent and individuals characteristics. Table O5. Saving choices and Mobile money. Robust standard errors vs. Standard errors. Table O6. Saving choices and mobile money: Low, irregular vs. High, regular incomes. Robust standard errors vs. Standard errors. Table O7. Saving choices and mobile money: Low vs. High access to formal financial instruments. Robust standard errors vs. Standard errors. Table O8. Usage and perception of mobile money and saving for health. Robust standard errors vs. Standard errors. Table O9. Receiving money transfers and mobile money. Full sample. Robust standard errors vs. Standard errors. 1

2 O1. External validity In the paper, we explain that our survey was conducted in the central region of Burkina Faso. Why was it chosen? How different is this region from the other Burkina s regions? In Burkina Faso there are 13 regions that consist of some urban where formal financial institutions are generally concentrated, and many rural ones 1 that are underserved or some without any formal financial institutions. Being the second region in terms of the number of mobile money agents, we chose the central region for budgetary reasons but also to increase our chance to find respondents who use mobile money services. Thus, our study allows us to analyze the s of using mobile money on people saving behavior where formal financial services are supposed to be available (urban areas) and also where they are supposed to be less developed or inexistent (rural areas). To provide some highlights about the similarities or differences between the 13 regions of Burkina Faso, we report data on formal financial institutions access by regions (Table A1). Data on individuals saving behavior and the number of mobile money accounts by region are not available. Where the data are available, we report the share of population and that of formal financial institutions located in each region (2012) as an indicator of access to formal financial services, and the number of mobile money agents (2014). Table A1 shows that central region is the most populated region with 13% of the population located in this area. Regarding the geographical breakdown of financial institutions, the data on banks are available only for the central region where about 36% of banks are located. Microfinance and credit union institutions are more concentrated in western part of the country (Boucle du Mouhoun) with respectively 21% and 16%. The available record on mobile money agents reveals that Boucle du Mouhoun has the highest number of mobile money agents with more than twice of that of the central region. O2. Impact of mobile money on individuals saving behavior using an alternative source of data. We use a survey data conducted on 1,000 people in Burkina Faso available in the Global Financial Inclusion Database (World Bank, 2015) in order to check the robustness of our main findings. While the database has the advantage of covering the whole country 2, it remains limited in providing individual-level characteristics such as location or type of income but allows us to replicate our core analysis. Table A2 presents the results. We use a logit model that mimics our equations (1) and (2) and two dependent variables for save for emergency and save to develop an. While the survey does not precisely identify savings for health, we define a proxy, save for emergency, indicating how individuals cope with an emergency. This proxy is a dummy 1 Generally, there is one urban area divided into many urban districts. For instance, in the central region, one urban area is divided into 12 urban districts, and 6 rural areas are divided into 172 rural districts. 2 Individual probability weights are used to make the sample nationally representative. 2

3 variable that equals to one if respondents indicate that it is very possible to come up with emergency funds through savings, and equals to zero otherwise. For the second dependent variable, save to develop an, we define a proxy that indicates if individuals save to start, operate, or grow a business or farm. This proxy is also a dummy that equals to one if respondents indicate that they saved to start, operate, or grow a business or farm, and equals to zero otherwise. We control for age, gender, education level, and income quintile. Due to lack of data, we only examine the heterogeneity of s of mobile money on individuals saving behavior by considering low vs. high income, female vs. male, and less vs. highly educated individuals. Overall, consistent with our findings, the results show that the use of mobile money increases the propensity of individuals to save for. The results also show that mobile money increases the propensity to save for especially for female and less educated individuals supporting our findings on disadvantaged groups. References The World Bank Group (2015) Data (online), The Global Financial Inclusion (Global Findex) Database, World Bank Group (2015). 3

4 Appendix Table O1. Access to formal financial services by region in Burkina Faso. Region Boucle du Mouhoun Cascades Centre Centre-Est Centre-Nord Centre-Ouest Centre-Sud Est Hauts-Bassins Nord Palreau-Central Sahel Population Banks Microfinance Credit union Mobile money agents % NA 21% 16% 238 4% NA 2% 5% NA 13% 36% 9% 12% 152 8% NA 4% 10% NA 8% NA 7% 5% NA 8% NA 10% 5% NA 4% NA 7% 2% NA 9% NA 5% 8% 34 11% NA 12% 13% NA 8% NA 9% 9% 43 5% NA 2% 3% NA 7% NA 10% 6% NA Sud-Ouest 4% NA 3% 6% NA Source: Ministère de l Economie et des Finances, The number of mobile money agents is provided by the mobile money provider Airtel money as of The terminology Central region used throughout the paper corresponds to the region Centre of this Table. 4

5 Appendix Table O2. Saving choices and mobile money: using Global Financial Inclusion Database. Full sample Low vs. High income Female vs. Male Less vs. Highly educated (1) (2) (3) (4) (5) (6) (7) MM user 0.945** 0.829** (0.378) (0.405) (0.450) (0.473) Individuals characteristics (1.705) (1.963) (1.956) MM user x Individuals characteristics ** ** (0.802) (0.692) (0.780) (0.637) (0.765) (0.601) Individuals characteristics x Controls included YES YES YES Observations Pseudo R *** 47.60*** 44.00*** 70.98*** Likelihood ratio test χ2 (H0: nullity of *** *** *** *** Full sample Low vs. High income Female vs. Male Less vs. Highly educated (1) (2) (3) (4) (5) (6) (7) MM user * 0.929** (0.473) (0.474) (0.586) (0.473) Individuals characteristics (2.367) (2.656) (4.245) MM user x Individuals characteristics 2.016** 1.583** (0.921) (0.789) (1.005) (0.816) (0.875) (0.736) Individuals characteristics x Controls included YES YES YES Observations Pseudo R *** 44.43*** 56.94*** 81.45*** Likelihood ratio test χ2 (H0: nullity of 82.24*** 84.66*** 87.40*** 83.71*** Note: Dependent variables: save for emergency and save to develop an are all dummies. Save for emergency equals to 1 if respondents indicate coming up with emergency funds through savings, and 0 otherwise. equals to 1 if respondents indicate save to start, operate, or grow business or farm, and 0 otherwise. The variable of interest, MM user is a binary variable that takes the value 1 if respondents has mobile money account, and 0 otherwise. To obtain the odds ratio, we simply compute the exponential of log odds. Robust standard errors are in brackets. Controls included: age, age squared, female, education level, income quintile and income quintile squared. According to the individual-level characteristics we remove respectively controls income quintile and income quintile squared, female and education level. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. 5

6 Appendix Table O3. Transaction fees of mobile money services. Fees associated with mobile money cash in / cash out functions and transfers services as of Mobile Money Services Minimum amount Maximum amount Fees (FCFA) Cash in (deposits) % ,20% ,50% Note: This payment system is a combination of a tiered/banded pricing and a percentage based pricing. Throughout, F CFA (Franc of the African Financial Community) refers to the local currency. The exchange rate during the survey period was about 500 F CFA = $1 US. 6

7 Appendix Table O4. Distance to the nearest mobile money agent and individuals characteristics. Distance to the nearest mobile money agent OLS Ordered Logit Coefficient RSE Coefficient RSE Age (0.009) (0.012) Age squared (0.000) (0.000) Married (0.133) (0.181) Rural * (0.133) (0.184) Male (0.134) (0.180) Occupation (0.179) (0.232) Irregular income (0.134) (0.181) At least one person in charge * (0.133) * (0.183) Education ** (0.070) ** (0.098) Income (0.097) (0.149) Income squared (0.017) (0.028) Note: Dependent variable: measure of agent access, takes value ranging from 1 to 5. RSE (robust standard errors) are in brackets. Each raw is a separate regression. We check the exogeneity of the distance to the nearest mobile money agent by examining whether it is correlated with individuals characteristics in our analysis and find only little evidence. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. 7

8 Appendix Table O5. Saving choices and Mobile money. Save for unpredictable purposes Save for health Save for predictable events Save to develop an Save for unpredictable events Logit regressions (standard errors) Save for health Save for predictable events Save to develop an (1) (2) (3) (4) (5) (6) (7) (8) User of MM 1.091** 0.922** ** 0.922** (0.527) (0.379) (0.324) (0.295) (0.527) (0.390) (0.315) (0.305) Age ** 0.438** * 0.438** (0.286) (0.261) (0.160) (0.181) (0.286) (0.196) (0.182) (0.174) Age squared ** ** (0.004) (0.004) (0.002) (0.003) (0.004) (0.003) (0.003) (0.002) Married *** ** (0.701) (0.550) (0.350) (0.356) (0.701) (0.457) (0.383) (0.383) Rural *** *** (0.438) (0.400) (0.322) (0.359) (0.438) (0.395) (0.322) (0.345) Male (0.505) (0.386) (0.318) (0.329) (0.505) (0.368) (0.304) (0.311) Occupation *** *** (0.998) (0.576) (0.542) (0.938) (0.998) (0.600) (0.513) (0.875) Irregular income *** 2.499*** *** 2.499*** (0.547) (0.451) (0.318) (0.376) (0.547) (0.424) (0.341) (0.361) Person in charge (0.411) (0.331) (0.291) (0.284) (0.411) (0.344) (0.291) (0.287) Education 0.629*** 0.382* *** 0.629*** *** (0.232) (0.208) (0.197) (0.185) (0.232) (0.234) (0.196) (0.183) Income (1.325) (1.402) (1.594) (1.222) (1.325) (1.399) (1.505) (1.122) Income squared * * (0.198) (0.223) (0.312) (0.198) (0.198) (0.242) (0.290) (0.194) Constant * ** ** (4.020) (3.801) (2.961) (3.242) (4.020) (3.292) (3.212) (3.141) Observations Pseudo R Wald χ2 (H0: nullity of 32.31*** 23.08** 40.96*** 71.39*** 32.31*** 19.21* 120.8*** 173.2*** Likelihood ratio test χ2 (H0: nullity 25.47** 24.62** 93.51*** *** 25.47** 24.62** *** *** of % correct prediction (y=1) 77.91% 52.26% 75.96% 85.79% 77.91% 52.26% 79.68% 85.79% % correct prediction (y=0) 61.54% 69.05% 73.85% 78.75% 61.54% 69.05% 75.49% 78.75% Note: Dependent variables: save for unpredictable purposes, save for health, save for predictable events and save to develop an are all dummies. Save for unpredictable purposes equal to 1 if respondents save for health and/or save for a potential decrease in income, and 0 otherwise. takes the value 1 if respondents indicate to save for health, and 0 otherw ise. Similarly, save for predictable events equal to 1 if respondents save to develop an or, save for education or, save to repay a loan and/or save for a ceremony (such as wedding or funeral), and 0 otherwise. also takes the value 1 if respondents save to develop an ac tivity, and o otherwise. The variable of interest, MM user is also a dummy that equal to 1 if respondents use mobile money, and 0 otherwise. The coefficients reported in the table are the log odds of the use of mobile money on saving patterns. To obtain the odds ratio, we simply compute the exponential of log odds. Robust standard errors and standard errors are in brackets respectively in columns 1 to 4 and columns 5 to 8. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. 8

9 Appendix Table O6. Saving choices and mobile money: Low, irregular vs. High, regular incomes. Logit regressions (standard errors) (1) (2) (3) (4) (5) (6) (7) (8) MM user 1.824*** *** (0.672) (0.484) (0.652) (0.474) Low income * * (6.788) (7.175) (6.843) (1, ) MM user x Low income * * * * * * (0.841) (0.505) (0.658) (0.447) (0.830) (0.513) (0.654) (0.450) Low income x Controls included YES YES YES YES Observations Pseudo R / 36.34** *** Likelihood ratio test χ2 (H0: nullity of 36.34** *** 30.93* 203.6*** % correct prediction (y=1) 86.77% 87.37% 86.77% 87.37% % correct prediction (y=0) 38.10% 76.25% 38.10% 76.25% Logit regressions (standard errors) (1) (2) (3) (4) (5) (6) (7) (8) MM user (0.539) (0.416) (0.547) (0.440) Irregular income * ** (7.845) (10.336) (8.472) (7.484) MM user x Irregular income 1.891** 2.092*** ** 2.092*** (0.907) (0.729) (0.677) (0.533) (0.889) (0.701) (0.671) (0.506) Irregular income x Controls included YES YES YES YES Observations Pseudo R * *** 45.34*** *** Likelihood ratio test χ2 (H0: nullity of 45.34*** *** 39.93** 207.4*** % correct prediction (y=1) 87.10% 87.89% 87.10% 87.89% % correct prediction (y=0) 57.14% 77.50% 57.14% 77.50% Note: Dependent variables: save for health and save to develop an. takes the value 1 if respondents indicate to save for health, and 0 otherwise. also takes the value 1 if respondents save to develop an, and o otherwise. The coefficients reported in the table are the log odds of the use of mobile money on saving patterns. To obtain the odds ratio, we simply compute the exponential of log odds. Robust standard errors and standard errors are in brackets respectively in columns 1 to 4 and columns 5 to 8. Low income individuals are those with less than 50,000 F CFA (around $100US) per month. Ir regular income individuals are those who specify having irregular income by answering the following question: Do you have regular or irregular income? The responses are encoded as irregular income = 1, and regular income = 0. Controls included: age, age squared, married, rural, male, occupation, irregular income, at least one person in charge, education level, income level and income squared. According to the individual-lev e l characteristics used we remove respectively controls income level and income squared, and irregular income. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. 9

10 Appendix Table O7. Saving choices and mobile money: Low vs. High access to formal financial instruments. Logit regressions (standard errors) (1) (2) (3) (4) (5) (6) (7) (8) MM user (0.544) (0.468) (0.552) (0.525) Rural *** (9.591) (8.445) (8.609) ( ) MM user x Rural * * (0.763) (0.535) (0.650) (0.451) (0.799) (0.578) (0.697) (0.458) Rural x Observations Pseudo R ** *** 33.30* *** Likelihood ratio test χ2 (H0: nullity of 33.30* *** *** % correct prediction (y=1) 84.84% 87.37% 84.84% 87.37% % correct prediction (y=0) 42.86% 79.38% 42.86% 79.38% Logit regressions (standard errors) (1) (2) (3) (4) (5) (6) (7) (8) MM user (0.580) (0.439) (0.579) (0.444) Female * *** * (6.827) (7.375) (7.947) ( ) MM user x Female 2.041** 2.024*** ** 2.024*** (0.881) (0.663) (0.644) (0.471) (0.857) (0.633) (0.657) (0.484) Female x Observations Pseudo R *** *** 43.31*** *** Likelihood ratio test χ2 (H0: nullity of 43.31*** *** 37.90** 187.2*** % correct prediction (y=1) 82.26% 86.32% 82.26% 86.32% % correct prediction (y=0) 47.62% 80.63% 47.62% 80.63% Logit regressions (standard errors) (1) (2) (3) (4) (5) (6) (7) (8) MM user (0.564) (0.430) (0.547) (0.446) Less educated ** ** (7.944) (7.649) ( ) ( ) MM user x Less educated ** ** (0.905) (0.708) (0.696) (0.547) (0.874) (0.682) (0.660) (0.486) Less educated x Observations Pseudo R *** *** 41.78*** *** Likelihood ratio test χ2 (H0: nullity of 41.78*** *** 37.13** 204.1*** % correct prediction (y=1) 88.50% 84.97% 88.50% 84.97% % correct prediction (y=0) 47.62% 81.88% 47.62% 81.88% Note: Dependent variables: save for health and save to develop an. takes the value 1 if respondents indicate to save for health, and 0 otherwise. also takes the value 1 if respondents save to develop an, and o otherwise. The coefficients reported in the table are the log odds of the use of mobile money on saving patterns. To obtain the odds ratio, we simply compute the exponential of log odds. Robust standard errors and standard errors are in brackets respectively in columns 1 to 4 and columns 5 to 8. Less educated individuals are those with primary education level or less (about six years of schooling at best). Controls included: age, age squared, married, rural, male, occupation, irregular income, at least one person in charge, education level, income level and income squared. According to the individual-level characteristics used we remove respectively controls rural, male and education level. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. 10

11 Appendix Table O8. Usage and perception of mobile money and saving for health. Safe place to make deposits Logit regressions (without robust standard errors) Transfer Transfer Low Transfers Increase Safe Low Transfers Increase within within cost of throughout mobile place to cost of throughout mobile the subregioregion the sub- money Burkina money make money Burkina money transfers Faso agent deposits transfers Faso agent (CI) (CI) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) MM user ** * (1.074) (1.648) (1.890) (1.001) (1.226) (1.130) (2.182) (1.902) (1.143) (1.420) Safe place to make deposits (6.949) (7.885) MM user x Safe place (1.230) (1.295) 1.347** 1.347** (0.600) (0.632) Low cost of money transfers (0.922) (1.112) MM user x Low cost (0.423) (0.550) 2.886** 2.886* (1.242) (1.646) Transfers throughout Burkina Faso (0.815) (1.039) MM user x Transfers throughout Burkina Faso (0.491) (0.491) (1.415) (1.427) Transfer within the sub-region (CI) (8.182) (8.332) MM user x Transfer within the sub-region (CI) (1.169) (1.296) 1.697*** 1.697*** (0.603) (0.610) Increase mobile money agent (0.863) (0.844) MM user x Increase mobile money agent (0.380) (0.419) (0.877) (1.028) YES YES YES YES YES YES Motivations x Controls included YES YES YES YES YES YES YES YES YES YES Observations Pseudo R *** 43.30** 39.17** 42.13** 43.45** 50.52*** 43.18** 38.05** 47.87*** 40.20** Likelihood ratio test χ2 (H0: nullity of 50.52*** 43.18** 38.05** 47.87*** 40.20** 44.85*** 37.26* ** 34.53* % correct prediction (y=1) 72.49% 72.40% 70.55% 84.47% 66.02% 72.49% 72.40% 70.55% 84.47% 66.02% % correct prediction (y=0) 71.43% 66.67% 64.29% 61.90% 71.43% 71.43% 66.67% 64.29% 61.90% 71.43% Note: Dependent variable:, is a dummy that takes the value 1 if respondents indicate to save for health, and 0 otherwise. Robust standard errors and standard errors are in brackets respectively in columns 1 to 5 and columns 6 to 10. Controls included: age, age squared, married, rural, male, occupation, irregular income, at least one person in charge, education level, income level and income squared. Table A.2 in the Appendix gives definitions and summary statistics of the independent variables. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. 11

12 Appendix Table O9. Receiving money transfers and mobile money. Full sample. Receiving money transfers Logit regression (robust standard errors) Logit regression (standard errors) (1) (2) MM user 1.828*** 1.828*** (0.287) (0.277) Age 0.351*** 0.351** (0.136) (0.146) Age squared *** ** (0.002) (0.002) Married (0.313) (0.333) Rural *** *** (0.310) (0.296) Male (0.261) (0.267) Occupation ** ** (0.523) (0.475) Irregular income 0.768** 0.768*** (0.313) (0.296) At least one person in charge (0.248) (0.252) Education (0.160) (0.163) Income 2.353** 2.353*** (0.914) (0.858) Income squared ** *** (0.150) (0.140) Constant *** *** (2.397) (2.419) Observations Pseudo R *** *** Likelihood ratio test χ2 (H0: nullity of *** 98.75*** % correct prediction (y=1) 74.90% 74.90% % correct prediction (y=0) 66.67% 66.67% Note: Dependent variable: Receiving money transfers is a dummy variable that equals 1 if respondents receive money transfers, and 0 otherwise. Robust standard errors and standard errors are in brackets respectively in column 1 and column 2. Controls included: age, age squared, married, rural, male, occupation, irregular income, at least one person in charge, education level, income level and income squared. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. 12